Date of Award

Spring 1994

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

First Advisor

Brown, Ronald H.

Second Advisor

Heinen, James A.

Third Advisor

Feng, Xin

Abstract

This thesis is concerned with the incorporation of a priori knowledge into feedforward artificial neural networks (ANNs). In most applications, an ANN is employed as a black box identification model, but the incorporation of any known information about the desired nonlinear mapping can improve identification. Various methods of incorporating a priori knowledge into ANNs have been proposed and one approach, the gray layer technique [5-9,29], is the focus of this thesis. The gray layer technique is examined in application to the identification of dynamic nonlinear systems. The technique is specifically applied to the identification of switched reluctance motor (SRM) state equations. The gray layer technique is compared to the conventional ANN architecture for both speed of convergence to the ANN estimate and on the basis of the ANNs ability to effectively identify the desired nonlinearity.

Share

COinS

Restricted Access Item

Having trouble?